this is the comind blog

devlog 2024-04-13

Added login tracking to the database and authentication server. I'm starting to think more about logging in general, since it'd be nice to have various metrics when I start letting users in. I do logging for some other stuff, like how many tokens go into/out of language models, which models are used, etc.

Shifting to a Julia-based websocket server for melds (Welder.jl). I watched a great YC video for technical founders, and Diana Hu's basic point was you should work with frameworks that help you with iteration speed, not things that scale. I can hack shit together in Julia very quickly. I realized that focusing too much on the Rust-based version of welder was slowing me down, so I cut it.

Expanded slug usage. Slugs are n-word permutations of one of 4,319 common English words. Slugs are assigned to thoughts and melds. Any thought or meld can be accessed via URL, i.e. They will also be displayed on UI elements. In principle, you can assign these codes by embedding similarity. Get the embedding for the new thought and cosine it against the slug embeddings.

I suspect you'd actually get this cool effect where slugs that seem most similar to the content are matched, like the thought

i left my keys in my car

might get you the slugs


or something to the equivalent.

I actually really like this idea.

Wrote the ripple blog post. I wrote a blog post about the ripple, an idea for the core Comind UX where each new thought "ripples" into 10-30 new thoughts.

The ripple was also interesting to think about from a core services perspective. Comind is not currently in the business of selling a front-end. It has a few frontends I've made, all open source.

None of these will require a purchase. Users can write their own frontend if they wish, since the service comind provides can be used in a bunch of different ways.

Basically, comind sells a service that lets you enter new thoughts and receive a ripple back. If you are a subscriber (not sure if that's the move right now) you get a higher quality ripple with custom thoughts from different cominds ({blog} to make blogs, {void} to talk about how the void is boring, etc.).

Free users can use all the communication tools, melds, etc. but won't have as many customized comind responses. Running a chat/forum is not terribly expensive, but LLMs are. We'll be carefully gating the LLM usage for free users because not doing so would bankrupt me.

Anyway, good session tonight!

Also – in heartening news, I got this email from the YCombinator founder matching platform:

Hey Cameron,

Hello from the Startup School and Co-Founder Matching team at YC!

We wanted to let you know that the deadline to apply for YC's Summer 2024 batch is coming up on 4/22.

We manually review everyone that signs up and we identified you as one of the top 10% on the site. If you're working on a startup, we'd strongly encourage you to apply to YC.

YC only runs twice a year, so if you're interested in doing YC for your startup, now is the time to apply.

You can apply here:

Best, The YC team

feels good man. have a good night y'all.

– Cameron

mindco © thanks to Franklin.jl and Julia.